CORRATA AGENT SHIELD
A list of AI apps isn’t control. Catch the agent while it’s working
AI agents now read email, open files and move data on your employees’ phones.
Corrata spots live agent sessions on iOS and Android, stops data heading to AI you didn’t approve, and does it all on the device.
No decryption. Nothing leaves the phone.
Microsoft MISA member
ISO 27001 certified
EU data residency
THE PROBLEM
An inventory tells you what’s installed. It doesn’t tell you a note-taking app just read 40 emails and sent them to a model.
The first wave of mobile AI security is discovery: a list of AI apps, sanctioned and not. Useful. But the risk isn’t the app. It’s what happens at 09:41 on a Tuesday when an agent decides to act.
01
AI hides inside ordinary apps
Keyboards, scanners, transcription and note apps call model APIs directly. They never show up as “AI apps” on a list.
02
Agents act at machine speed
An agent can read a mailbox, open a file and upload it in under two seconds. A weekly report finds out after the data has gone.
03
“Capable” isn’t “active”
Flagging apps that could act as agents creates a list of suspects. Security teams need to know which one is acting, right now, and stop it.
Discovery vs control
Most tools stop at the list. We start there.
Question |
A discovery tool answers |
Corrata Agent Shield answers |
|---|---|---|
What AI is here? |
The AI apps installed on each device |
Every app that talks to a model, including AI buried inside ordinary apps |
Is an agent active? |
Which apps have permissions to act |
Which session is running now, scored Human, Assisted, or Autonomous |
What do we do? |
Alert, report, update policy |
Pause the session, block the upload, ask the user, cut Microsoft 365 access |
Can we prove it? |
Usage reports |
An Agent Receipt for every autonomous session |
CAPABILITIES
Six controls built for how agents actually behave.
LIVE AGENT DETECTION
Spot the session, not just the app
Agents leave a pattern: rapid loops between a model and your services, actions while the screen is off, work no thumb could do. We read that pattern from traffic metadata and score every session Human, Assisted or Autonomous.
HIDDEN AI EXPOSURE
We don’t hunt for AI SDKs. We see the calls.
Static scans miss AI added in a silent update. Corrata watches where each app sends data, so any app that starts talking to a model shows up the moment it does.
SESSION PAUSE
“Is this you?”
When an autonomous session reaches for email, files or finance apps, Corrata can pause it and ask the user on their phone. One tap to continue. No tap, no data.
AI DATA GUARDRAILS
An approved lane for AI. A wall for the rest.
Let Copilot and your approved tools work. Block uploads to unapproved AI services, or cap them by size, app and user group. Powered by Corrata Advanced DLP.
AGENT RECEIPTS
Every agent leaves a receipt
Each autonomous session gets a timeline: which app, which AI service, which company systems it touched, how much data moved, what we did about it. Ready for your auditor, your DPO and EU AI Act evidence.
ACCESS THAT REACTS
AI risk becomes device risk
A device running an unapproved agent can be marked non-compliant in Microsoft Intune. Conditional access cuts Microsoft 365 until it’s resolved. No new console for your team to learn.
HOW IT WORKS
See. Attribute. Judge. Act. All on the phone.
STEP 1: SEE
Corrata inspects every connection on the device before it leaves. We match destinations against a live map of model APIs, AI apps, agent platforms and MCP servers.
STEP 2: ATTRIBUTE
Each flow is tied to the app that made it. Not “someone used AI”. This app, this user, this much data.
STEP 3: JUDGE
ming, sequence and volume are combined with device signals like accessibility and overlay permissions. The result is a session score, in milliseconds.
STEP 4: ACT
Allow, pause, block or ask. Every decision is logged as an Agent Receipt and can flow to your SIEM and Intune.
PRIVACY BY ARCHITECTURE
Governing AI shouldn’t mean reading your people’s messages.
Built in Dublin for European rules. Agent Shield works on metadata, on the device. It knows an agent moved 2 MB to an unapproved service. It never knows what the 2 MB said.
No decryption
No certificates to install, no TLS interception, no pinning breakage.
Inspection stays on the phone
Traffic is analysed locally. Only verdicts and receipts reach the console.
EU data residency, ISO 27001
Your AI governance data stays in the EU.
Built for works council and DPO review
BYOD friendly. Personal app traffic can be excluded by policy.
STRAIGHT TALK
What no mobile security tool can see. Including ours.
Any vendor claiming total AI visibility on mobile is overselling. Here’s where the line sits, so you can plan around it.
Full on-device models
AI that runs entirely on the chip sends no traffic. We flag the apps that use it, but can’t watch the thinking.
Inside the iOS sandbox
Apple doesn’t let any third party watch one app driving another. iOS detection relies on traffic, and is strongest on supervised devices.
Platform assistants
Siri and Gemini run partly on Apple and Google infrastructure. We govern them through MDM policy, not by pretending to intercept them.
Cloud agents
Agents running in a remote browser never touch the phone. That’s an identity and bot problem, and we’ll tell you which tools solve it.
SEE IT FOR YOURSELF
We’ll run an AI agent on a test phone. You watch Corrata catch it.
Live session, live verdict, live receipt. Bring your toughest question about mobile AI. We’ll show you the answer on a real device.
Partner or MSSP? Agent Shield is built to resell. Talk to our channel team.